FakeScope: from image forensics to election fact-checking

The report visits the Recod.ai team to explore AI-generated election content. The coverage concerns the wider collaboration; a named FakeScope appearance has not been verified. According to the laboratory’s recap, Mateus de Pádua Vicente demonstrates detection tools using a suspected candidate video, and Anderson Rocha discusses the potential influence of synthetic media on voters.

Read Recod.ai’s recap (5 October)

The programme includes Anderson Rocha presenting Unicamp’s deepfake-detection research, bringing together lighting, texture, and pixel clues. The Recod.ai communications team identified this programme as part of its publicity for FakeScope.

The laboratory’s communications team announced plans for TV Cultura to feature FakeScope. The report will be added here once a public programme or article link is available.

Juliana Vicentini’s report explains how FakeScope combines a quantitative assessment with explanations of visual clues, including anatomy, lighting, environmental interactions, and text. It quotes Yixuan Li on the need for verifiable evidence and describes the model’s role in VigIA’s election monitoring.

The article describes FakeScope’s visual-evidence analysis and its complementary quantitative and qualitative assessments. It quotes Yixuan Li on explainability and reports the tool’s use within VigIA, which brings together Agência Lupa, Recod.ai, FAAP, and Meedan.

This feature presents FakeScope within the Horus project and explains its ability to describe why an image may be synthetic. It reports a practical example involving a fabricated video of Lebanon’s president, in which FakeScope identified distortions in a flag and around image edges.

This fact-check reports that a video was split into 458 frames for analysis by Mateus de Pádua Vicente using FakeScope. The resulting forensic explanations identified issues including unnaturally uniform light and distorted building perspective, providing evidence for the verification.

A dedicated feature publicizing FakeScope’s approach to explainable image forensics. The laboratory’s accompanying announcement presents it as a digital detective that explains suspicious visual evidence and credits Yixuan Li, Anderson Rocha, and collaborators in China and Singapore.

The post introduces FakeScope’s ability to explain the clues behind its decisions, linking this transparency to efforts against disinformation and fraud. It names Yixuan Li at the University of Bristol and Anderson Rocha at Unicamp, alongside international collaborators.

Lupa’s methodology documents the use of FakeScope alongside Pixel-Inconsistency and Detect and Locate. FakeScope adds explanations of visual and forensic clues to support the assessment. Analysts combine the tools’ evidence with journalistic context and human review; the page also cites our paper.